Table 7 Ablation study of different pruning methods in ET module on three datasets.
From: Multi-branch CNN and grouping cascade attention for medical image classification
Id | Method | Params (M) | FLOPs (G) | Acc | F1 | Precision | Recall | Auc |
|---|---|---|---|---|---|---|---|---|
BUSI | ||||||||
(0) | Baseline | 45.8 | 9.9 | 0.9133 | 0.8964 | 0.9102 | 0.8869 | 0.9162 |
(1) | (0)+ET Module | 53.2 | 11.6 | 0.9297 | 0.9259 | 0.9295 | 0.9226 | 0.9386 |
(2) | (1)+EC Module | 25.5 | 6.4 | 0.9333 | 0.9261 | 0.9326 | 0.9226 | 0.9404 |
COVID19-CT | ||||||||
(0) | Baseline | 45.8 | 9.9 | 0.8986 | 0.9057 | 0.9000 | 0.9114 | 0.8977 |
(1) | (0)+ET Module | 53.2 | 11.6 | 0.9257 | 0.9308 | 0.9250 | 0.9367 | 0.9249 |
(2) | (1)+EC Module | 25.5 | 6.4 | 0.9257 | 0.9317 | 0.9146 | 0.9494 | 0.9240 |
Chaoyang | ||||||||
(0) | Baseline | 45.8 | 9.9 | 0.8504 | 0.7940 | 0.8019 | 0.7887 | 0.8689 |
(1) | (0)+ET Module | 53.2 | 11.6 | 0.8565 | 0.8066 | 0.8229 | 0.7952 | 0.8729 |
(2) | (1)+EC Module | 25.5 | 6.4 | 0.8635 | 0.8090 | 0.8191 | 0.8012 | 0.8776 |
